FreeBSD/Linux Kernel Cross Reference
sys/geom/geom_dev.c
1 /*-
2 * Copyright (c) 2002 Poul-Henning Kamp
3 * Copyright (c) 2002 Networks Associates Technology, Inc.
4 * All rights reserved.
5 *
6 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7 * and NAI Labs, the Security Research Division of Network Associates, Inc.
8 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 * DARPA CHATS research program.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The names of the authors may not be used to endorse or promote
20 * products derived from this software without specific prior written
21 * permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: releng/10.1/sys/geom/geom_dev.c 260385 2014-01-07 01:32:23Z scottl $");
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/malloc.h>
42 #include <sys/kernel.h>
43 #include <sys/conf.h>
44 #include <sys/ctype.h>
45 #include <sys/bio.h>
46 #include <sys/bus.h>
47 #include <sys/lock.h>
48 #include <sys/mutex.h>
49 #include <sys/proc.h>
50 #include <sys/errno.h>
51 #include <sys/time.h>
52 #include <sys/disk.h>
53 #include <sys/fcntl.h>
54 #include <sys/limits.h>
55 #include <sys/sysctl.h>
56 #include <geom/geom.h>
57 #include <geom/geom_int.h>
58 #include <machine/stdarg.h>
59
60 struct g_dev_softc {
61 struct mtx sc_mtx;
62 struct cdev *sc_dev;
63 struct cdev *sc_alias;
64 int sc_open;
65 int sc_active;
66 };
67
68 static d_open_t g_dev_open;
69 static d_close_t g_dev_close;
70 static d_strategy_t g_dev_strategy;
71 static d_ioctl_t g_dev_ioctl;
72
73 static struct cdevsw g_dev_cdevsw = {
74 .d_version = D_VERSION,
75 .d_open = g_dev_open,
76 .d_close = g_dev_close,
77 .d_read = physread,
78 .d_write = physwrite,
79 .d_ioctl = g_dev_ioctl,
80 .d_strategy = g_dev_strategy,
81 .d_name = "g_dev",
82 .d_flags = D_DISK | D_TRACKCLOSE,
83 };
84
85 static g_taste_t g_dev_taste;
86 static g_orphan_t g_dev_orphan;
87 static g_attrchanged_t g_dev_attrchanged;
88
89 static struct g_class g_dev_class = {
90 .name = "DEV",
91 .version = G_VERSION,
92 .taste = g_dev_taste,
93 .orphan = g_dev_orphan,
94 .attrchanged = g_dev_attrchanged
95 };
96
97 /*
98 * We target 262144 (8 x 32768) sectors by default as this significantly
99 * increases the throughput on commonly used SSD's with a marginal
100 * increase in non-interruptible request latency.
101 */
102 static uint64_t g_dev_del_max_sectors = 262144;
103 SYSCTL_DECL(_kern_geom);
104 SYSCTL_NODE(_kern_geom, OID_AUTO, dev, CTLFLAG_RW, 0, "GEOM_DEV stuff");
105 SYSCTL_QUAD(_kern_geom_dev, OID_AUTO, delete_max_sectors, CTLFLAG_RW,
106 &g_dev_del_max_sectors, 0, "Maximum number of sectors in a single "
107 "delete request sent to the provider. Larger requests are chunked "
108 "so they can be interrupted. (0 = disable chunking)");
109
110 static void
111 g_dev_destroy(void *arg, int flags __unused)
112 {
113 struct g_consumer *cp;
114 struct g_geom *gp;
115 struct g_dev_softc *sc;
116
117 g_topology_assert();
118 cp = arg;
119 gp = cp->geom;
120 sc = cp->private;
121 g_trace(G_T_TOPOLOGY, "g_dev_destroy(%p(%s))", cp, gp->name);
122 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
123 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
124 g_detach(cp);
125 g_destroy_consumer(cp);
126 g_destroy_geom(gp);
127 mtx_destroy(&sc->sc_mtx);
128 g_free(sc);
129 }
130
131 void
132 g_dev_print(void)
133 {
134 struct g_geom *gp;
135 char const *p = "";
136
137 LIST_FOREACH(gp, &g_dev_class.geom, geom) {
138 printf("%s%s", p, gp->name);
139 p = " ";
140 }
141 printf("\n");
142 }
143
144 static void
145 g_dev_attrchanged(struct g_consumer *cp, const char *attr)
146 {
147 struct g_dev_softc *sc;
148 struct cdev *dev;
149 char buf[SPECNAMELEN + 6];
150
151 sc = cp->private;
152 if (strcmp(attr, "GEOM::media") == 0) {
153 dev = sc->sc_dev;
154 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
155 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf, M_WAITOK);
156 dev = sc->sc_alias;
157 if (dev != NULL) {
158 snprintf(buf, sizeof(buf), "cdev=%s", dev->si_name);
159 devctl_notify_f("DEVFS", "CDEV", "MEDIACHANGE", buf,
160 M_WAITOK);
161 }
162 return;
163 }
164
165 if (strcmp(attr, "GEOM::physpath") != 0)
166 return;
167
168 if (g_access(cp, 1, 0, 0) == 0) {
169 char *physpath;
170 int error, physpath_len;
171
172 physpath_len = MAXPATHLEN;
173 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
174 error =
175 g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
176 g_access(cp, -1, 0, 0);
177 if (error == 0 && strlen(physpath) != 0) {
178 struct cdev *old_alias_dev;
179 struct cdev **alias_devp;
180
181 dev = sc->sc_dev;
182 old_alias_dev = sc->sc_alias;
183 alias_devp = (struct cdev **)&sc->sc_alias;
184 make_dev_physpath_alias(MAKEDEV_WAITOK, alias_devp,
185 dev, old_alias_dev, physpath);
186 } else if (sc->sc_alias) {
187 destroy_dev((struct cdev *)sc->sc_alias);
188 sc->sc_alias = NULL;
189 }
190 g_free(physpath);
191 }
192 }
193
194 struct g_provider *
195 g_dev_getprovider(struct cdev *dev)
196 {
197 struct g_consumer *cp;
198
199 g_topology_assert();
200 if (dev == NULL)
201 return (NULL);
202 if (dev->si_devsw != &g_dev_cdevsw)
203 return (NULL);
204 cp = dev->si_drv2;
205 return (cp->provider);
206 }
207
208 static struct g_geom *
209 g_dev_taste(struct g_class *mp, struct g_provider *pp, int insist __unused)
210 {
211 struct g_geom *gp;
212 struct g_consumer *cp;
213 struct g_dev_softc *sc;
214 int error, len;
215 struct cdev *dev, *adev;
216 char buf[64], *val;
217
218 g_trace(G_T_TOPOLOGY, "dev_taste(%s,%s)", mp->name, pp->name);
219 g_topology_assert();
220 gp = g_new_geomf(mp, "%s", pp->name);
221 sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
222 mtx_init(&sc->sc_mtx, "g_dev", NULL, MTX_DEF);
223 cp = g_new_consumer(gp);
224 cp->private = sc;
225 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
226 error = g_attach(cp, pp);
227 KASSERT(error == 0,
228 ("g_dev_taste(%s) failed to g_attach, err=%d", pp->name, error));
229 error = make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK, &dev,
230 &g_dev_cdevsw, NULL, UID_ROOT, GID_OPERATOR, 0640, "%s", gp->name);
231 if (error != 0) {
232 printf("%s: make_dev_p() failed (gp->name=%s, error=%d)\n",
233 __func__, gp->name, error);
234 g_detach(cp);
235 g_destroy_consumer(cp);
236 g_destroy_geom(gp);
237 mtx_destroy(&sc->sc_mtx);
238 g_free(sc);
239 return (NULL);
240 }
241 dev->si_flags |= SI_UNMAPPED;
242 sc->sc_dev = dev;
243
244 /* Search for device alias name and create it if found. */
245 adev = NULL;
246 for (len = MIN(strlen(gp->name), sizeof(buf) - 15); len > 0; len--) {
247 snprintf(buf, sizeof(buf), "kern.devalias.%s", gp->name);
248 buf[14 + len] = 0;
249 val = getenv(buf);
250 if (val != NULL) {
251 snprintf(buf, sizeof(buf), "%s%s",
252 val, gp->name + len);
253 freeenv(val);
254 make_dev_alias_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
255 &adev, dev, "%s", buf);
256 adev->si_flags |= SI_UNMAPPED;
257 break;
258 }
259 }
260
261 dev->si_iosize_max = MAXPHYS;
262 dev->si_drv2 = cp;
263 if (adev != NULL) {
264 adev->si_iosize_max = MAXPHYS;
265 adev->si_drv2 = cp;
266 }
267
268 g_dev_attrchanged(cp, "GEOM::physpath");
269
270 return (gp);
271 }
272
273 static int
274 g_dev_open(struct cdev *dev, int flags, int fmt, struct thread *td)
275 {
276 struct g_consumer *cp;
277 struct g_dev_softc *sc;
278 int error, r, w, e;
279
280 cp = dev->si_drv2;
281 if (cp == NULL)
282 return(ENXIO); /* g_dev_taste() not done yet */
283 g_trace(G_T_ACCESS, "g_dev_open(%s, %d, %d, %p)",
284 cp->geom->name, flags, fmt, td);
285
286 r = flags & FREAD ? 1 : 0;
287 w = flags & FWRITE ? 1 : 0;
288 #ifdef notyet
289 e = flags & O_EXCL ? 1 : 0;
290 #else
291 e = 0;
292 #endif
293 if (w) {
294 /*
295 * When running in very secure mode, do not allow
296 * opens for writing of any disks.
297 */
298 error = securelevel_ge(td->td_ucred, 2);
299 if (error)
300 return (error);
301 }
302 g_topology_lock();
303 error = g_access(cp, r, w, e);
304 g_topology_unlock();
305 if (error == 0) {
306 sc = cp->private;
307 mtx_lock(&sc->sc_mtx);
308 if (sc->sc_open == 0 && sc->sc_active != 0)
309 wakeup(&sc->sc_active);
310 sc->sc_open += r + w + e;
311 mtx_unlock(&sc->sc_mtx);
312 }
313 return(error);
314 }
315
316 static int
317 g_dev_close(struct cdev *dev, int flags, int fmt, struct thread *td)
318 {
319 struct g_consumer *cp;
320 struct g_dev_softc *sc;
321 int error, r, w, e;
322
323 cp = dev->si_drv2;
324 if (cp == NULL)
325 return(ENXIO);
326 g_trace(G_T_ACCESS, "g_dev_close(%s, %d, %d, %p)",
327 cp->geom->name, flags, fmt, td);
328
329 r = flags & FREAD ? -1 : 0;
330 w = flags & FWRITE ? -1 : 0;
331 #ifdef notyet
332 e = flags & O_EXCL ? -1 : 0;
333 #else
334 e = 0;
335 #endif
336 sc = cp->private;
337 mtx_lock(&sc->sc_mtx);
338 sc->sc_open += r + w + e;
339 while (sc->sc_open == 0 && sc->sc_active != 0)
340 msleep(&sc->sc_active, &sc->sc_mtx, 0, "PRIBIO", 0);
341 mtx_unlock(&sc->sc_mtx);
342 g_topology_lock();
343 error = g_access(cp, r, w, e);
344 g_topology_unlock();
345 return (error);
346 }
347
348 /*
349 * XXX: Until we have unmessed the ioctl situation, there is a race against
350 * XXX: a concurrent orphanization. We cannot close it by holding topology
351 * XXX: since that would prevent us from doing our job, and stalling events
352 * XXX: will break (actually: stall) the BSD disklabel hacks.
353 */
354 static int
355 g_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
356 {
357 struct g_consumer *cp;
358 struct g_provider *pp;
359 struct g_kerneldump kd;
360 off_t offset, length, chunk;
361 int i, error;
362 u_int u;
363
364 cp = dev->si_drv2;
365 pp = cp->provider;
366
367 error = 0;
368 KASSERT(cp->acr || cp->acw,
369 ("Consumer with zero access count in g_dev_ioctl"));
370
371 i = IOCPARM_LEN(cmd);
372 switch (cmd) {
373 case DIOCGSECTORSIZE:
374 *(u_int *)data = cp->provider->sectorsize;
375 if (*(u_int *)data == 0)
376 error = ENOENT;
377 break;
378 case DIOCGMEDIASIZE:
379 *(off_t *)data = cp->provider->mediasize;
380 if (*(off_t *)data == 0)
381 error = ENOENT;
382 break;
383 case DIOCGFWSECTORS:
384 error = g_io_getattr("GEOM::fwsectors", cp, &i, data);
385 if (error == 0 && *(u_int *)data == 0)
386 error = ENOENT;
387 break;
388 case DIOCGFWHEADS:
389 error = g_io_getattr("GEOM::fwheads", cp, &i, data);
390 if (error == 0 && *(u_int *)data == 0)
391 error = ENOENT;
392 break;
393 case DIOCGFRONTSTUFF:
394 error = g_io_getattr("GEOM::frontstuff", cp, &i, data);
395 break;
396 case DIOCSKERNELDUMP:
397 u = *((u_int *)data);
398 if (!u) {
399 set_dumper(NULL, NULL);
400 error = 0;
401 break;
402 }
403 kd.offset = 0;
404 kd.length = OFF_MAX;
405 i = sizeof kd;
406 error = g_io_getattr("GEOM::kerneldump", cp, &i, &kd);
407 if (!error) {
408 error = set_dumper(&kd.di, devtoname(dev));
409 if (!error)
410 dev->si_flags |= SI_DUMPDEV;
411 }
412 break;
413 case DIOCGFLUSH:
414 error = g_io_flush(cp);
415 break;
416 case DIOCGDELETE:
417 offset = ((off_t *)data)[0];
418 length = ((off_t *)data)[1];
419 if ((offset % cp->provider->sectorsize) != 0 ||
420 (length % cp->provider->sectorsize) != 0 || length <= 0) {
421 printf("%s: offset=%jd length=%jd\n", __func__, offset,
422 length);
423 error = EINVAL;
424 break;
425 }
426 while (length > 0) {
427 chunk = length;
428 if (g_dev_del_max_sectors != 0 && chunk >
429 g_dev_del_max_sectors * cp->provider->sectorsize) {
430 chunk = g_dev_del_max_sectors *
431 cp->provider->sectorsize;
432 }
433 error = g_delete_data(cp, offset, chunk);
434 length -= chunk;
435 offset += chunk;
436 if (error)
437 break;
438 /*
439 * Since the request size can be large, the service
440 * time can be is likewise. We make this ioctl
441 * interruptible by checking for signals for each bio.
442 */
443 if (SIGPENDING(td))
444 break;
445 }
446 break;
447 case DIOCGIDENT:
448 error = g_io_getattr("GEOM::ident", cp, &i, data);
449 break;
450 case DIOCGPROVIDERNAME:
451 if (pp == NULL)
452 return (ENOENT);
453 strlcpy(data, pp->name, i);
454 break;
455 case DIOCGSTRIPESIZE:
456 *(off_t *)data = cp->provider->stripesize;
457 break;
458 case DIOCGSTRIPEOFFSET:
459 *(off_t *)data = cp->provider->stripeoffset;
460 break;
461 case DIOCGPHYSPATH:
462 error = g_io_getattr("GEOM::physpath", cp, &i, data);
463 if (error == 0 && *(char *)data == '\0')
464 error = ENOENT;
465 break;
466 default:
467 if (cp->provider->geom->ioctl != NULL) {
468 error = cp->provider->geom->ioctl(cp->provider, cmd, data, fflag, td);
469 } else {
470 error = ENOIOCTL;
471 }
472 }
473
474 return (error);
475 }
476
477 static void
478 g_dev_done(struct bio *bp2)
479 {
480 struct g_consumer *cp;
481 struct g_dev_softc *sc;
482 struct bio *bp;
483 int destroy;
484
485 cp = bp2->bio_from;
486 sc = cp->private;
487 bp = bp2->bio_parent;
488 bp->bio_error = bp2->bio_error;
489 bp->bio_completed = bp2->bio_completed;
490 bp->bio_resid = bp->bio_length - bp2->bio_completed;
491 if (bp2->bio_error != 0) {
492 g_trace(G_T_BIO, "g_dev_done(%p) had error %d",
493 bp2, bp2->bio_error);
494 bp->bio_flags |= BIO_ERROR;
495 } else {
496 g_trace(G_T_BIO, "g_dev_done(%p/%p) resid %ld completed %jd",
497 bp2, bp, bp2->bio_resid, (intmax_t)bp2->bio_completed);
498 }
499 g_destroy_bio(bp2);
500 destroy = 0;
501 mtx_lock(&sc->sc_mtx);
502 if ((--sc->sc_active) == 0) {
503 if (sc->sc_open == 0)
504 wakeup(&sc->sc_active);
505 if (sc->sc_dev == NULL)
506 destroy = 1;
507 }
508 mtx_unlock(&sc->sc_mtx);
509 if (destroy)
510 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
511 biodone(bp);
512 }
513
514 static void
515 g_dev_strategy(struct bio *bp)
516 {
517 struct g_consumer *cp;
518 struct bio *bp2;
519 struct cdev *dev;
520 struct g_dev_softc *sc;
521
522 KASSERT(bp->bio_cmd == BIO_READ ||
523 bp->bio_cmd == BIO_WRITE ||
524 bp->bio_cmd == BIO_DELETE ||
525 bp->bio_cmd == BIO_FLUSH,
526 ("Wrong bio_cmd bio=%p cmd=%d", bp, bp->bio_cmd));
527 dev = bp->bio_dev;
528 cp = dev->si_drv2;
529 sc = cp->private;
530 KASSERT(cp->acr || cp->acw,
531 ("Consumer with zero access count in g_dev_strategy"));
532 #ifdef INVARIANTS
533 if ((bp->bio_offset % cp->provider->sectorsize) != 0 ||
534 (bp->bio_bcount % cp->provider->sectorsize) != 0) {
535 bp->bio_resid = bp->bio_bcount;
536 biofinish(bp, NULL, EINVAL);
537 return;
538 }
539 #endif
540 mtx_lock(&sc->sc_mtx);
541 KASSERT(sc->sc_open > 0, ("Closed device in g_dev_strategy"));
542 sc->sc_active++;
543 mtx_unlock(&sc->sc_mtx);
544
545 for (;;) {
546 /*
547 * XXX: This is not an ideal solution, but I belive it to
548 * XXX: deadlock safe, all things considered.
549 */
550 bp2 = g_clone_bio(bp);
551 if (bp2 != NULL)
552 break;
553 pause("gdstrat", hz / 10);
554 }
555 KASSERT(bp2 != NULL, ("XXX: ENOMEM in a bad place"));
556 bp2->bio_done = g_dev_done;
557 g_trace(G_T_BIO,
558 "g_dev_strategy(%p/%p) offset %jd length %jd data %p cmd %d",
559 bp, bp2, (intmax_t)bp->bio_offset, (intmax_t)bp2->bio_length,
560 bp2->bio_data, bp2->bio_cmd);
561 g_io_request(bp2, cp);
562 KASSERT(cp->acr || cp->acw,
563 ("g_dev_strategy raced with g_dev_close and lost"));
564
565 }
566
567 /*
568 * g_dev_callback()
569 *
570 * Called by devfs when asynchronous device destruction is completed.
571 * - Mark that we have no attached device any more.
572 * - If there are no outstanding requests, schedule geom destruction.
573 * Otherwise destruction will be scheduled later by g_dev_done().
574 */
575
576 static void
577 g_dev_callback(void *arg)
578 {
579 struct g_consumer *cp;
580 struct g_dev_softc *sc;
581 int destroy;
582
583 cp = arg;
584 sc = cp->private;
585 g_trace(G_T_TOPOLOGY, "g_dev_callback(%p(%s))", cp, cp->geom->name);
586
587 mtx_lock(&sc->sc_mtx);
588 sc->sc_dev = NULL;
589 sc->sc_alias = NULL;
590 destroy = (sc->sc_active == 0);
591 mtx_unlock(&sc->sc_mtx);
592 if (destroy)
593 g_post_event(g_dev_destroy, cp, M_WAITOK, NULL);
594 }
595
596 /*
597 * g_dev_orphan()
598 *
599 * Called from below when the provider orphaned us.
600 * - Clear any dump settings.
601 * - Request asynchronous device destruction to prevent any more requests
602 * from coming in. The provider is already marked with an error, so
603 * anything which comes in in the interrim will be returned immediately.
604 */
605
606 static void
607 g_dev_orphan(struct g_consumer *cp)
608 {
609 struct cdev *dev;
610 struct g_dev_softc *sc;
611
612 g_topology_assert();
613 sc = cp->private;
614 dev = sc->sc_dev;
615 g_trace(G_T_TOPOLOGY, "g_dev_orphan(%p(%s))", cp, cp->geom->name);
616
617 /* Reset any dump-area set on this device */
618 if (dev->si_flags & SI_DUMPDEV)
619 set_dumper(NULL, NULL);
620
621 /* Destroy the struct cdev *so we get no more requests */
622 destroy_dev_sched_cb(dev, g_dev_callback, cp);
623 }
624
625 DECLARE_GEOM_CLASS(g_dev_class, g_dev);
Cache object: 07603dc99467c84f9093e11254934018
|